Biomedical Engineering Reference
In-Depth Information
Table 4.
Comparison Between Manual and Algorithm Segmentation
for the Six Prostates (Permission granted by the AAPM)
Volume (cm
3
)
Prostate
k
∆
V
k
%
MD
k
(mm)
MAD
k
(mm)
MAXD
k
(mm)
Manual
V
m,k
Algorithm
V
a,k
1
36.93
33.63
8.95
-0.45
1.16
8.45
2
27.36
26.33
3.75
-0.10
1.03
5.69
3
22.87
20.80
9.05
-0.32
1.39
7.04
4
24.22
22.42
7.41
0.13
1.03
6.59
5
23.54
20.85
11.42
-0.48
1.33
7.58
6
21.58
21.06
2.39
-0.12
1.17
7.32
Average
26.08
24.18
7.16
-0.20
1.19
7.01
Standard
Deviation 5.65
5.08
3.45
0.28
0.14
1.04
Reprinted with permission from the AAPM.
Table 4 lists the volumes calculated from themanually and algorithm-segmented
boundaries for each prostate
k
as well as
V
k
%. The volumes calculated from the
manually segmented (i.e., “gold standard”) boundaries are larger than those calcu-
lated from algorithm-segmented boundaries, and this is reflected in positive values
for
V
k
%. This table also shows the global metrics MD, MAD, and MAXD for
each prostate as well as their averages and standard deviations. It is not clear
whether the differences between the two methods are caused by inaccuracies in
semiautomatic segmentation, or by manual segmentation, and it is not possible to
resolve this in the absence of a true gold standard.
Figures 19a,b show the local standard deviation in the manually segmented
meshes that were repeatedly segmented from the same 3D image by the same tech-
nician. For clarity, the local standard deviation is mapped on top of the average
manually segmented mesh. Figure 19a is a view that is perpendicular to the trans-
verse plane in the direction from the base of the prostate to the apex, and Figure 19b
is a view perpendicular to the sagittal plane in the direction from the patient's right
to left. Figures 19c,d show the same views as in Figures 19a,b, respectively, but for
the algorithm-segmented meshes. The average standard deviation (averaged over
all vertices) for the manual segmentation was 0
.
98
±
0
.
01 mm, whereas the average
standard deviation was lower for the algorithm, with a value of 0
.
63
±
0
.
02 mm.
The local variability in algorithm-segmented boundaries is lowest in regions where
the prostate boundary is clearly visible. In these regions, automatic deformation
can guide vertices toward the prostate boundary even when there is variability in
